Let $M \subseteq \mathbb{R}^n$ be bounded and $N_{\epsilon}(M)$ the minimum number of 'squares' of side $\epsilon$ with center in M necessary to cover $M$. The box dimension of M is then defined as $\dim_B \; M = \lim\limits_{\epsilon \to 0+} \frac{\log \; N_{\epsilon}(M)}{\log \; \epsilon^{-1}}$. The following proposition shows that the s-dimensional fractal content $N_{\epsilon}(M) \epsilon^s$ presents a $0-\infty$-behaviour similar to the s-dimensional Hausdorff measure:


1.$\lim\limits_{\epsilon \to 0+} N_{\epsilon}(M) \epsilon^s = \infty$ if $s < dim_B M$

2.$\lim\limits_{\epsilon \to 0+} N_{\epsilon}(M) \epsilon^s = 0$ if $s > dim_B M$

In argument 1, I show the existence of an element $s_0$ where the above behaviour happens. That is, I prove the proposition replacing $dim_B M$ by $s_0$.

The difference with the Hausdorff dimension definition is that in that setting we defined the dimension as the jumping value in the observed in the $0-\infty$-behaviour. Here, I need to show that $s_0 = dim_B M$.

Until now I only have an informal argument for the case $dim_B M \notin \{0,\infty\}$. So:


How can I formally show that $s_0 = dim_B M$?

Argument 1

For $t > s \ge 0$ write $N_{\epsilon}(M) \epsilon^t = \epsilon^{t-s} N_{\epsilon}(M) \epsilon^s$. If $\lim\limits_{\epsilon \to 0+} N_{\epsilon}(M) \epsilon^s < \infty$ then $\lim\limits_{\epsilon \to 0+} N_{\epsilon}(M) \epsilon^t = 0$. On the other hand, if $\lim\limits_{\epsilon \to 0+} N_{\epsilon}(M) \epsilon^t < \infty$ and $\neq 0$ then $\lim\limits_{\epsilon \to 0+} N_{\epsilon}(M) \epsilon^s = \infty$. So we can confirm a $0-\infty$-behaviour as shown in the diagram:

enter image description here


Falconer, Fractal Geometry: Mathematical Foundations and Applications, 2nd edition, page 46. Elgar, Measure, Topology and Fractal Geometry, 2nd edition, page 213.

  • $\begingroup$ Do you explicitly assume that the limit $\dim_B \; M = \lim_{\epsilon \to 0+} \frac{\log \; N_{\epsilon}(M)}{\log \; \epsilon^{-1}}$ exists? The lower box dimension may be strictly less than the upper box dimension. See also Falconer's book on p. 42 (2.8). $\endgroup$ – Skeeve Apr 3 at 19:37
  • $\begingroup$ @Skeeve From my understanding of Falconer, yes, $dim_B \; M$ is assumed to exist and from its limit definition one should prove that it is the critical value for the fractal content. $\endgroup$ – Javier Apr 3 at 19:42
  • $\begingroup$ By the way, your Argument 1 shows only that there exist $\underline s$ and $\overline s$ such that if $s> \overline s$ then $N_\epsilon \epsilon^{s} \to 0$ as $\epsilon \to 0$, and if $s < \underline s$ then $N_\epsilon \epsilon^s \to \infty$ as $\epsilon\to0$. But it is not evident that $\underline s = \overline s$. The point is that your Argument 1 does not seem to use that the limit in the definition of $\dim_B M$ exists. $\endgroup$ – Skeeve Apr 3 at 21:00

Since $M$ is a bounded subset of $\mathbb R^n$ it automatically holds that $N_\epsilon \le \frac{C}{\epsilon^{-n}}$, hence $d=\dim_B M \le n$.

By the definition of limit for any $\delta>0$ there exists $\epsilon_0>0$ such that for all $\epsilon \in (0,\epsilon_0)$ it holds $d-\delta < \frac{\log N_\epsilon}{\log \epsilon^{-1}} < d+\delta$, hence $\epsilon^{-(d-\delta)} < N_\epsilon < \epsilon^{-(d+\delta)}$. So if $s > d+\delta$ then $N_\epsilon \epsilon^{s} \to 0$ as $\epsilon\to 0$. Similarly, if $0< s < d-\delta$ then $N_\epsilon \epsilon^{s} \to \infty$ as $\epsilon \to 0$. This argument shows that $s_0 \in [d-\delta, d+\delta]$ and by arbitrariness of $\delta>0$ we conclude that $s_0 = \delta$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.